Gemma 3 27B
Gemma 3 27B needs 18.1 GB of VRAM at Q4_K_M with 8k context, computed from its config.json as checked on 16 October 2026. The smallest card that holds it has 24 GB, and the fastest desktop card that runs it, the RTX 5090, generates about 79 tokens per second.
- Parameters
- 27.4B
- Context
- 131,072 tokens
- Licence
- Gemma Terms of Use
- Released
- March 2025
- Config checked
- 16 October 2026
Open Gemma 3 27B in the LLM checker
Which card size runs it
18.1 GB against each memory size. A model may use 95% of a card, and only what is left over, or missing, is coloured; tight means above 85%, with little room for a longer conversation.
Other quantisations, context lengths or a Q8 KV cache: the VRAM calculator.
Computed 16 October 2026Which GPU to pick
The smallest card here that loads it at Q4_K_M with 8k of context is the RX 7900 XT (20 GB): 18.1 GB needed, an estimated 35 tokens per second, and a tight fit that leaves little for the conversation. For room to work, meaning 32,768 tokens of context while staying under 85% of memory, step up to the RTX 3090 (24 GB), which needs 20.1 GB for that and generates at about 41 tokens per second.
At Q8_0, which is near-lossless, the card is the RTX A6000 (48 GB), needing 31.8 GB. On Apple silicon, a Mac mini M4 Pro holds it with 32,768 tokens of context at an estimated 12 tokens per second; prompt processing is slower than on Nvidia, which long documents make obvious. Among unified-memory desktops, the Ryzen AI Max+ 395 holds it at 32,768 tokens in 20.1 GB, at an estimated 11 tokens per second.
Speed on each card
| GPU | Fits | Est. tokens/s | Feels like | Memory | Bandwidth |
|---|---|---|---|---|---|
| H100 SXM | Yes | ≤ 148 | faster than you can read | 80 GB | 3,350 GB/s |
| A100 | Yes | 90 | faster than you can read | 80 GB | 2,039 GB/s |
| RTX 5090 | Yes | 79 | faster than you can read | 32 GB | 1,792 GB/s |
| RTX PRO 6000 Blackwell | Yes | 79 | faster than you can read | 96 GB | 1,792 GB/s |
| RTX PRO 5000 Blackwell | Yes | 59 | comfortable for chat | 48 GB | 1,344 GB/s |
| RTX 4090 | Yes | 44 | comfortable for chat | 24 GB | 1,008 GB/s |
| RX 7900 XTX | Yes | 42 | comfortable for chat | 24 GB | 960 GB/s |
| RTX 6000 Ada | Yes | 42 | comfortable for chat | 48 GB | 960 GB/s |
| RTX 3090 | Yes | 41 | comfortable for chat | 24 GB | 936 GB/s |
| RTX PRO 4500 Blackwell | Yes | 39 | comfortable for chat | 32 GB | 896 GB/s |
| L40S | Yes | 38 | comfortable for chat | 48 GB | 864 GB/s |
| RX 7900 XT | Tight | 35 | comfortable for chat | 20 GB | 800 GB/s |
| RTX A6000 | Yes | 34 | comfortable for chat | 48 GB | 768 GB/s |
| RTX PRO 4000 Blackwell | Yes | 30 | comfortable for chat | 24 GB | 672 GB/s |
| Radeon AI PRO R9700 | Yes | 28 | comfortable for chat | 32 GB | 640 GB/s |
| Arc Pro B70 | Yes | 27 | comfortable for chat | 32 GB | 608 GB/s |
| Arc Pro B60 | Yes | 20 | usable, a little slow | 24 GB | 456 GB/s |
| RTX PRO 4000 Blackwell SFF | Yes | 19 | usable, a little slow | 24 GB | 432 GB/s |
| RTX Spark | Yes | 13 | usable, a little slow | 128 GB (102 for the GPU) | 300 GB/s |
| DGX Spark | Yes | 12 | usable, a little slow | 128 GB (126 for the GPU) | 273 GB/s |
| Ryzen AI Max+ 395 | Yes | 11 | slow; fine for batch jobs | 128 GB (96 for the GPU) | 256 GB/s |
| RTX 4060 Ti | No | - | does not fit | 16 GB | 288 GB/s |
| RTX 5060 Ti | No | - | does not fit | 16 GB | 448 GB/s |
| RTX 5080 | No | - | does not fit | 16 GB | 960 GB/s |
| RTX 5070 Ti | No | - | does not fit | 16 GB | 896 GB/s |
| RTX 4080 SUPER | No | - | does not fit | 16 GB | 736 GB/s |
| RX 9070 XT | No | - | does not fit | 16 GB | 640 GB/s |
| RX 9060 XT | No | - | does not fit | 16 GB | 320 GB/s |
| RTX 3060 | No | - | does not fit | 12 GB | 360 GB/s |
| RTX 4070 | No | - | does not fit | 12 GB | 504 GB/s |
| RTX 5070 | No | - | does not fit | 12 GB | 672 GB/s |
| Arc B580 | No | - | does not fit | 12 GB | 456 GB/s |
Single-stream decode, from memory bandwidth at 70% efficiency. Prompt processing and batching not included. These are ceilings, not measurements: Gemma 3 27B reads only 15.9 GB of weights per token at Q4, so little that the per-token costs the formula leaves out decide much of the real speed. The figure marked ≤ is where a real runtime falls furthest below the number shown. See the speed estimator for other quantisations and Apple hardware, or every card compared if you are choosing hardware rather than a model.
Computed 16 October 2026Memory by quantisation and context
| Quant | 4,096 ctx | 8,192 ctx | 32,768 ctx |
|---|---|---|---|
| Q8_0 | 31.5 GB · fits 48 GB | 31.8 GB · fits 48 GB | 33.8 GB · fits 48 GB |
| Q5_K_M | 21.5 GB · fits 24 GB | 21.8 GB · fits 24 GB | 23.9 GB · fits 32 GB |
| Q4_K_M | 17.8 GB · fits 24 GB | 18.1 GB · fits 24 GB | 20.1 GB · fits 24 GB |
Gemma 4 31B is the newer model in this family. It needs about 20.9 GB at the same setting.
Weights at this quant: 15.9 GB at Q4. Every extra 1,000 tokens of context adds about 0.082 GB of KV cache at FP16, on top of a fixed 0.436 GB that the sliding-window layers hold once their window is full. Try other settings in the VRAM calculator.
About the model
Gemma 3 27B has 27.4B parameters. It has 62 layers with 16 KV heads of dimension 128; five in every six use sliding-window attention and only remember the last 1,024 tokens, so 10 global layers carry the long context. Google's largest Gemma 3 model, March 2025: vision-language, 128k window, and for a year the most common answer to "what do I run on a 24 GB card".
Text and images 131,072-token window Gemma Terms of Use March 2025
Reasoning. Answers directly; there is no reasoning mode.
A runtime without a sliding-window cache stores every layer at full length and needs far more than this at long context.
Native context window: 131,072 tokens. Licence: Gemma Terms of Use, a custom licence with its own conditions; read it before commercial use. Weights published in March 2025.
Good at
- A 24 GB card, including long context, because five layers in six only remember 1,024 tokens.
- Multilingual and image work with a mature set of quantisations.
Watch out for
- Gemma Terms of Use, not Apache.
- Gemma 4 31B replaces it with a longer window, a thinking mode and an Apache licence.
Similar models
The models that need about the same memory as Gemma 3 27B at Q4_K_M and 8k, which makes them the real alternatives on whatever card you have. The context column is what each extra 1,000 tokens costs, and it is where models of the same size differ most.
| Model | Size | Needs | Per 1k context | Window | Released |
|---|---|---|---|---|---|
| Gemma 3 27B | 27.4B | 18.1 GB | 0.082 GB | 128k | 2025-03 |
| Qwen3.8 27B | 27.8B | 18 GB | 0.066 GB | 256k | 2026-08 |
| Qwen3.6 27B | 27.8B | 18 GB | 0.066 GB | 256k | 2026-04 |
| Muse Glimmer 30B | 29.8B | 18.7 GB | 0.013 GB | 128k | 2026-08 |
| Nemotron 3.5 Lightning 30B-A3B (MoE) | 31.6B · 3B active | 19.7 GB | 0.006 GB | 256k | 2026-08 |
See every model compared, which models fit on 8 to 96 GB, or the cheapest hour of every cloud card.
Questions
- Can I run Gemma 3 27B on a 24 GB card?
- Yes, at Q4_K_M and 8k context it needs about 18.1 GB, which fits a 24 GB card with room.
- How much VRAM does Gemma 3 27B need at Q8?
- About 31.8 GB at 8k context, or 33.8 GB at 32k. Q8 is near-lossless; use it when it fits.
- What is the context length of Gemma 3 27B?
- 131,072 tokens natively. Holding all of it at Q4_K_M takes about 28.2 GB, of which 11.2 GB is context.
- Why does Gemma 3 27B need so little memory for long context?
- Because most of its layers do not keep a cache that grows. It has 62 layers with 16 KV heads of dimension 128; five in every six use sliding-window attention and only remember the last 1,024 tokens, so 10 global layers carry the long context. At 32k tokens the context costs about 3.1 GB; if all 62 layers kept a conventional full-length cache, the same conversation would cost about 16.6 GB. That saving depends on the runtime implementing the sliding-window cache. llama.cpp and vLLM do; a build that does not will use far more.
- Is there a newer model than Gemma 3 27B?
- Yes: Gemma 4 31B, published in March 2026. At the same Q4_K_M and 8k setting it needs about 20.9 GB against 18.1 GB for this one.
- How fast is Gemma 3 27B on an RTX 4090?
- At most about 44 tokens per second at Q4, single stream, which is comfortable for chat. That is a ceiling worked out from the card's memory bandwidth, not a measurement.
- Architecture values from config.json in google/gemma-3-27b-it on Hugging Face (the repository is licence-gated: the card is public, the file opens once you accept the terms), checked 16 October 2026. Every figure on this page is computed from those values; check them before buying hardware for this model.
- Estimates assume a single conversation on a card that is otherwise free. A desktop on the same GPU takes 0.5 to 2 GB.